ER residency of the ceramide phosphoethanolamine synthase SMSr relies on homotypic oligomerization mediated by its SAM domain

SMSr/SAMD8 is an ER-resident ceramide phosphoethanolamine synthase with a critical role in controlling ER ceramides and suppressing ceramide-induced apoptosis in cultured cells. SMSr-mediated ceramide homeostasis relies on the enzyme’s catalytic activity as well as on its N-terminal sterile α-motif or SAM domain. Here we report that SMSr-SAM is structurally and functionally related to the SAM domain of diacylglycerol kinase DGKδ, a central regulator of lipid signaling at the plasma membrane. Native gel electrophoresis indicates that both SAM domains form homotypic oligomers. Chemical crosslinking studies show that SMSr self-associates into ER-resident trimers and hexamers that resemble the helical oligomers formed by DGKδ-SAM. Residues critical for DGKδ-SAM oligomerization are conserved in SMSr-SAM and their substitution causes a dissociation of SMSr oligomers as well as a partial redistribution of the enzyme to the Golgi. Conversely, treatment of cells with curcumin, a drug disrupting ceramide and Ca2+ homeostasis in the ER, stabilizes SMSr oligomers and promotes retention of the enzyme in the ER. Our data provide first demonstration of a multi-pass membrane protein that undergoes homotypic oligomerization via its SAM domain and indicate that SAM-mediated self-assembly of SMSr is required for efficient retention of the enzyme in the ER.


SMSr-SAM lacks affinity for lipids.
We previously demonstrated that SMSr is a CPE synthase that requires its N-terminal SAM domain to control ceramide levels in the ER 25 . As SAM domains have been reported to bind lipids 11,26,27 including sphingolipids 12 , we performed photo-affinity labeling experiments with isolated SMSr-SAM produced in E. coli using bifunctional lipid analogues that contain a photo-activatable diazirine and a clickable alkyne group (Fig. 1a,b) 28 . Proteins in direct contact with a bifunctional lipid can be crosslinked by UV irradiation of the diazirine group. Next, click chemistry is used to label the alkyne group with a fluorescent reporter, allowing visualization of the crosslinked protein-lipid complex by in-gel fluorescence. As shown in Fig. 1c, the recombinant START domain of the ceramide transfer protein CERT could be specifically crosslinked with bifunctional ceramide. In contrast, recombinant SMSr-SAM was devoid of any specific lipid binding affinity when subjected to photo-crosslinking with bifunctional analogues of ceramide (Cer), CPE, SM, DAG, phosphatidylethanolamine (PE) or phosphatidylcholine (PC). This led us to explore alternative functions of SMSr-SAM than lipid binding.

SMSr-SAM is structurally and functionally related to DGKδ-SAM. A BLAST search with
SMSr-SAM as query yielded the SAM domain of DAG kinase DGKδ as a hit with the lowest Expect (E) value (i.e. 3E-05) (Suppl. Table 1; Fig. 2a,b). In addition, phylogenetic analysis revealed that SMSr-SAM is more closely related to DGKδ -SAM than to the SAM domain of SMS1 (Fig. 2c). Isolated DGKδ -SAM has been shown to self-assemble into helical oligomers through a head-to-tail interaction, with six SAM monomers per turn 7 (Fig. 2d). SAM-mediated oligomerization of DGKδ controls its function as a key regulator of lipid signaling by sequestering the enzyme in an inactive cellular location 6 . Moreover, several key residues involved in DGKδ -SAM homo-oligomerization are conserved in SMSr-SAM from human to zebrafish, but do not occur in SMS1-SAM ( Fig. 2b; residues marked by asterisks). This suggested that SMSr-SAM and DGKδ -SAM share a similar function.
To investigate whether SMSr-SAM, like DGKδ -SAM, is able to self-associate into oligomers, we used a method originally developed by Knight et al. 29 . To this end, SMSr-SAM and DGKδ -SAM were expressed in bacteria as fusions to a super-negatively charged GFP (scGFP) and run on a native gel (Fig. 2e). The negative charge on GFP minimizes the chance of GFP-mediated self-assembly and ensures that all proteins migrate to the cathode while the green fluorescence can be used to monitor the migration of the fusion proteins in the native gel. As shown in Fig. 2f, polymeric scGFP-DGKδ -SAM displayed a major shift in migration relative to its monomeric control, scGFP-DGKδ -SAM V52E . In comparison to scGFP-DGKδ -SAM, scGFP-SMSr-SAM migrated even slower through the gel while a significant portion of the protein remained in the gel well (data not shown). These results are consistent with those reported by Knight et al. 29 and suggested that scGFP-SMSr-SAM forms stable oligomers. To verify that oligomerization is mediated by the SAM domain, we mutated each of three conserved residues that were previously shown to be critical for homo-oligomerization of DGKδ -SAM, yielding SMSr-SAM L62E , SMSr-SAM G63D and SMSr-SAM K66E (Fig. 2b). All three mutants migrated faster through the gel than wild-type SMSr-SAM (Fig. 2g), indicative of a defect in homo-oligomerization. From this we conclude that SMSr-SAM is able to form homo-oligomers.
Scientific RepoRts | 7:41290 | DOI: 10.1038/srep41290 SMSr forms SAM-dependent oligomers in the ER. To investigate whether SMSr forms homo-oligomers in cells, we performed co-immunoprecipitation (IP) experiments on HeLa cells co-expressing V5-and HA-tagged versions of the enzyme. Immunoblot analysis of anti-V5 immunoprecipitates revealed, besides monomeric SMSr, also higher-order enzyme complexes that were resistant to SDS under non-reducing conditions (Fig. 3a). The migration profiles of the SDS-resistant enzyme complexes and their cross-reactivity with both anti-V5 and anti-HA antibodies suggested that they correspond to SMSr trimers and hexamers (see also below). When immunoprecipitations were performed on HeLa cells co-expressing full-length SMSr-HA and a truncated version of SMSr-V5 lacking the N-terminal SAM-domain, the anti-V5 immunoprecipitates were devoid of SMSr-HA and only contained SMSrΔ SAM monomers and dimers (Fig. 3a). Thus, the ability of SMSr to form trimers and hexamers appeared critically dependent on its SAM domain. We previously reported that removal of its SAM domain causes SMSr to redistribute from the ER to the Golgi 25 (see also below). To address the possibility that immunoprecipitation of SMSrΔ SAM-V5 fails to bring down SMSr-HA because both proteins reside in different organelles, SMSrΔ SAM-V5 was retained in the ER by adding a KKXX ER-retrieval motif to its C-terminus. As shown in Fig. 3a, anti-SMSrΔ SAM-V5-KKSA immunoprecipitates were still devoid of SMSr-HA. In contrast, co-immunoprecipitation analysis on cells co-expressing full-length and SAM-truncated versions of SMS1 revealed that this enzyme self-associates independently of its SAM domain (Fig. 3b). Thus, consistent with its ability to self-associate in vitro, SMSr-SAM appears to drive homo-oligomerization of SMSr in the ER.
We noticed that under non-reducing conditions, SMSr monomers, trimers and hexamers all migrate as discrete smears in the gel. Treatment of immunoprecipitates with DTT prior to gel electrophoresis dissolved most of the trimers and hexamers, and caused the monomer to migrate as a single protein band (Fig. 3c), indicating that Recombinantly produced SAM domains are incubated with liposomes containing bifunctional lipid analogues and subjected to UV irradiation. Click chemistry is used to label the alkyne group in the bifunctional lipid with N 3 -AlexaFluor647, allowing visualization of UV-crosslinked protein-lipid complexes by in-gel fluorescence. (c) Recombinant GFP-fusions of the SAM domains of SMSr and diacylglycerol kinase DGKδ were subjected to lipid photoaffinity labeling using bifunctional analogues of ceramide (pacCer), diacylglycerol (pacDAG), CPE (pacCPE), phosphoethanolamine (pacPE), sphingomyelin (pacSM) and phosphatidylcholine (pacPC), processed for SDS-PAGE and analyzed by in-gel fluorescence (right) then stained with Coomassie (left). The ceramide-binding domain of ceramide transfer protein CERT (CERT-START) and a corresponding ceramidebinding mutant (CERT-START N504A ) served as controls.
Scientific RepoRts | 7:41290 | DOI: 10.1038/srep41290  Figure 3. SMSr forms SAM-dependent homo-oligomers in the ER. (a) Detergent extracts of HeLa cells co-transfected with HA-tagged SMSr and V5/His6-tagged SMSr, SMSr∆ SAM, SMSr-KKSA or SMSr∆ SAM-KKSA were subjected to immunoprecipitation analysis using an anti-V5 antibody. Immunoprecipitates (IP) and total extracts (input) were immunoblotted (IB) using anti-V5 and anti-HA antibodies. An asterisk denotes immunoreactivity with IgG heavy chain. (b) Detergent extracts of HeLa cells co-transfected with HA-tagged SMS1 and V5/His6-tagged SMS1 or SMS1∆ SAM were subjected to immunoprecipitation analysis using an anti-V5 antibody. Immunoprecipitates (IP) and total extracts (input) were immunoblotted (IB) using anti-V5 and anti-HA antibodies, as in (a). (c) HeLa cells transfected with empty vector (EV) or V5/His6-tagged SMSr were solubilized using detergent in the presence or absence of 10 mM N-ethyl maleimide (NEM) and subjected to Ni 2+ -NTA affinity chromatography. Ni 2+ -NTA eluates were incubated in the presence or absence of 100 mM DTT, and immunoblotted (IB) using an anti-V5 antibody, as in (a). the electrophoretic mobility of SMSr is strongly influenced by the presence of both intermolecular and intramolecular disulfide bonds. To investigate whether these disulfide bonds form before or after cell lysis, cells were lysed in the presence of N-ethylmaleimide (NEM), a membrane permeable agent that reacts with free thiol groups. NEM treatment recapitulated the effect of DTT addition (Fig. 3c), indicating that the intermolecular and intramolecular disulfide bonds in SMSr form after cell lysis, presumably due to the oxidative milieu provided by the lysis buffer. Henceforth, NEM was included in the lysis buffer in all subsequent experiments. To verify that SMSr also forms trimers and hexamers in intact cells, HeLa cells expressing SMSr-HA were treated with the membrane permeable and amine-reactive crosslinker DSP prior to immunopreciptation analysis. As shown in Fig. 3d, DSP treatment led to the appearance of SMSr trimers and hexamers in anti-HA immunoprecipitates. These trimers and hexamers became more abundant at the expense of monomers when increasing amounts of crosslinker were used. SMSr trimers and hexamers could also be recovered from immunoprecipitates of DSP-treated and SMSr-V5-expressing Saccharomyces cerevisiae, an organism lacking endogenous SMSr homologs. Interestingly, some trimers and hexamers were also observed when crosslinker was omitted (Fig. 3e). In contrast, immunoprecipitates prepared from DSP-treated HeLa cells expressing SMSrΔ SAM-V5 contained monomers and dimers but were devoid of any trimers or hexamers (Fig. 3f). From this we conclude that SMSr-SAM mediates self-assembly of SMSr into trimers and hexamers in the ER. SMSr lacking its N-terminal SAM-domain, on the other hand, appears to retain the ability to form dimers.

Isolation and functional analysis of oligomerization-defective SMSr mutants.
To study the functional relevance of SMSr oligomerization in HeLa cells, we first generated an oligomerization-defective SMSr mutant. IP analysis of V5-tagged SMSr mutants carrying one of three single residue substitutions that blocked SMSr-SAM oligomerization in vitro (Fig. 2g) revealed that each of these reduced the ability of the enzyme to self-associate into trimers and hexamers (Fig. 4a). When all three single residue substitutions were combined, we observed a substantial reduction in the self-associating properties of SMSr as judged by co-IP analysis (Fig. 4b).
We named this oligomerization-defective triple mutant SMSr OD . To allow a detailed functional analysis of SMSr OD without interference from any endogenous pool of oligomerization-competent SMSr, we created HeLa SMSr knockout (SMSr −/− ) cells using CRISPR/Cas9 technology. Contrary to wild-type cells, SMSr −/− cells lack a 39-kDa protein that cross-reacts with a well-characterized anti-SMSr antibody 22 (Fig. 4c). Moreover, SMSr −/− cells were virtually devoid of CPE synthase activity (Fig. 4d,e). This is consistent with previous RNAi experiments indicating that SMSr is the principal CPE synthase in HeLa cells 21 . The residual CPE synthase activity detected in SMSr −/− cells is likely due to SMS2, which serves as a bifunctional enzyme with both SM and CPE synthase activity 30 . Unlike HeLa cells treated with SMSr-targeting siRNAs, SMSr −/− cells lack any signs of apoptosis (our unpublished data). Whether this is due to a compensatory mechanism that overcomes a deregulation of ER ceramides over time remains to be established. As shown in Fig. 4f, SMSr OD displayed a reduced ability to self-associate into hexamers and trimers following its immunoprecipitation from either DSP-treated wild-type or SMSr −/− cells. Besides a marked decrease in the recovery of hexamers and trimmers, DSP-crosslinking of SMSr OD led to appearance of dimers (Fig. 4f). Dimers were also observed in crosslinking experiments with the SAM-domain truncation mutant but not with the wild-type enzyme (Fig. 3f). Together, these results provide complementary evidence that SAM-mediated self-assembly is the key mechanism by which SMSr forms trimers and hexamers in the ER.

SMSr oligomerization is dispensable for catalytic activity and vice versa.
To address whether an impaired capacity to form homo-oligomers influences the catalytic activity of SMSr, we expressed V5-tagged SMSr, SMSr OD and a catalytically dead SMSr D348E mutant in Saccharomyces cerevisiae, an organism lacking endogenous CPE synthase activity. Next, catalytic activities of the heterologously expressed enzymes were determined in cell lysates using a recently established quantitative CPE synthase activity assay 31 . Whereas the D348E mutation completely abolished SMSr-mediated CPE production, there was no significant difference between the catalytic activities of SMSr and SMSr OD , indicating that the oligomeric state of the enzyme does not critically influence its catalytic capacity (Fig. 5a,b). We then addressed whether disrupting catalytic activity influences the ability of SMSr to form homo-oligomers. Although introduction of the D348E mutation appeared to render SMSr less stable when expressed in HeLa cells, co-IP analysis of SMSr −/− cells co-expressing V5-and HA-tagged SMSr D348E showed that catalytic activity per se is not a prerequisite for SMSr self-assembly (Fig. 5c). Moreover, DSP crosslinking experiments revealed that SMSr D348E displays the same capacity to self-assemble into hexamers and trimers as the wild-type enzyme (Fig. 5d). Collectively, these results indicate that the ability of SMSr to form homo-oligomers is dispensable for its catalytic activity and vice versa. Curcumin promotes SMSr oligomerization. As SMSr is a critical regulator of ER ceramides 21,25 and because ceramides can trigger ER stress by disrupting Ca 2+ homeostasis 32 , we wondered whether an acute imbalance in ER ceramide or Ca 2+ levels would influence the oligomeric state of the enzyme. Several drugs have been reported to acutely alter ceramide and/or Ca 2+ levels in the ER. For instance, curcumin stimulates de novo ceramide biosynthesis, presumably by directly activating ceramide synthases in the ER [33][34][35] . Tamoxifen has been reported to increase ceramide levels by blocking glucosylceramide biosynthesis 36,37 whereas thapsigargin is known to release Ca 2+ from the ER lumen by inhibiting the sarco/endoplasmic reticulum Ca 2+ -ATPase, SERCA 38 . As shown in Fig. 6a, treatment of SMSr-V5-expressing HeLa cells with curcumin led to the appearance of higher-order enzyme complexes in anti-V5 immunoprecipitates that were resistant to SDS-PAGE under reducing conditions. No such complexes were observed in immunoprecipitates from control cells or from cells treated with tamoxifen or thapsigargin. Curcumin treatment did not trigger formation of higher-order complexes of calnexin, an abundant membrane-anchored chaperone of the ER. The curcumin-induced SMSr complexes co-migrated, at least in part, with DSP-crosslinked SMSr trimers and hexamers (Fig. 6b), indicating that curcumin promotes the formation of SDS-resistant SMSr homo-oligomers. A distinct SMSr-containing complex of approximately 110 kDa was consistently and exclusively found in curcumin-treated cells (marked with an asterisk in Fig. 6a,b). As this curcumin-induced complex did not match the migration profiles of SMSr dimers (~85 kDa), trimers (~125 kDa) or hexamers (~250 kDa), we anticipate that it represents a heterologous complex comprising SMSr and another, yet-to-be identified protein. Identification of this SMSr interaction partner is the subject of ongoing investigations.
Curcumin-induced SMSr oligomerization is independent of fluctuations in ER ceramide or calcium levels. As curcumin has been reported to trigger mitochondrial apoptosis 33,35,39 , we first examined whether its ability to promote SMSr homo-oligomerization relies on activation of an apoptotic pathway. A time course experiment revealed that curcumin triggers SMSr oligomerization already within one hour (Fig. 6c). Moreover, treatment of cells with the pan-caspase inhibitor zVAD-fmk blocked curcumin-induced PARP cleavage but did not prevent formation of SDS-resistant SMSr oligomers (Fig. 6d). Together, these results indicate that curcumin promotes SMSr homo-oligomerization independently of its apoptogenic activity. Since curcumin stimulates de novo ceramide biosynthesis, we next investigated whether formation of SDS-resistant SMSr oligomers in curcumin-treated cells is due to an accumulation of ceramides in the ER. We previously reported that over-expression of the ceramide transfer protein CERT rescues SMSr-depleted cells from ceramide-induced apoptosis by stimulating ceramide export from the ER 25   A heterologous SMSr-protein complex of ~100 kDa formed in curcumin-treated cells is marked by an asterisk. Uncropped images of blots are provided in Supplementary Fig. S4. over-expression did not prevent formation of SDS-resistant SMSr oligomers in curcumin-treated cells. We then used various well-characterized drugs that block either de novo ceramide biosynthesis or ceramide export from the ER. As shown in Fig. 6f, long chain base synthase inhibitor myriocin or ceramide synthase inhibitor fumonisin B1 in each case had no impact on curcumin-induced SMSr oligomerization when added 5 h prior to curcumin treatment. Also, addition of HPA-12, a specific inhibitor of CERT-mediated ceramide transport 40 , did not lead to any obvious changes in the oligomeric state of SMSr in either control or curcumin-treated cells (Fig. 6f). From this we conclude that curcumin-induced oligomerization of SMSr occurs independently of fluctuations in ER ceramide levels.
Besides deregulating ceramide levels in the ER, curcumin promotes the release of Ca 2+ from the ER lumen by inhibiting SERCA 41 . BAPTA-AM, a membrane permeable Ca 2+ chelator, has been reported to block some of the downstream effects of curcumin 42,43 . However, pre-treatment of cells with BAPTA-AM did not prevent curcumin-induced SMSr oligomerization (Fig. 6g). Moreover, addition of thapsigargin, a specific inhibitor of SERCA, did not mimic the effect of curcumin on SMSr oligomerization. This indicates that curcumin-induced formation of SDS-resistant SMSr oligomers is unlikely mediated by a perturbed Ca 2+ homeostasis in the ER. Curcumin has previously been reported to crosslink the cystic fibrosis transmembrane conductance receptor into SDS-resistant oligomers 44 and to promote dimerization of ceramide synthases 34 . Thus, the appearance of SDS-resistant SMSr oligomers in cells treated with curcumin may be due to the crosslinking activity of this drug.
SMSr oligomerization is critical for ER localization. The subcellular distribution of DGKδ is controlled by its SAM domain, which prevents translocation of the enzyme to the plasma membrane by mediating its self-assembly into cytosolic oligomers 6 . Removal of its N-terminal SAM domain caused SMSr to redistribute from the ER to the Golgi complex 21 (Fig. 7a). In contrast, removal of the N-terminal SAM domain from SMS1 did not affect its Golgi residency (Fig. 7b). Swapping the entire N-terminal cytoplasmic tail of SMS1 against that of SMSr, on the other hand, caused the protein to be retained in the ER 21 . To investigate whether SAM-mediated homo-oligomerization of SMSr plays a role in retaining the enzyme in the ER, we created chimeric SMS1 proteins in which the SAM domain is swapped for SMSr-derived wild-type or oligomerization-defective SAM. We named these chimeric proteins SAMr-SMS1 and SAMr OD -SMS1, respectively. As shown in Fig. 7b, both SAMr-SMS1 and SAMr OD -SMS1 exclusively localized to the Golgi complex. Collectively, these results indicated that SMSr-SAM does not mediate ER retention or, alternatively, that such an activity is obscured by the presence of an ER export signal in the N-terminal cytoplasmic tail of SMS1 located downstream of its SAM domain. The latter scenario is supported by our finding that addition of a C-terminal KKXX ER-retrieval sequence caused a redistribution of SMSrΔ SAM from the Golgi to the ER, but failed to retain SMS1 in the ER (Fig. 7a,b).
As alternative approach to investigate whether SAM-mediated oligomerization is part of the mechanism by which SMSr is retained in the ER, we next analyzed the subcellular distribution of GFP-tagged SMSr, SMSr∆ SAM and SMSr OD in HeLa SMSr −/− cells using confocal fluorescence microscopy. As shown in Fig. 8a, GFP-SMSr primarily resides in the ER while a minor but significant pool of the enzyme is localized in the Golgi complex. Indeed, intensity plots along a line that crosssections the Golgi complex demonstrated partially overlapping intensity peaks for GFP-SMSr and the Golgi marker GM130; such overlap was not observed between intensity plots for GM130 and the ER marker calnexin (Fig. 8b). In contrast to GFP-SMSr and in line with our previous findings, the bulk of GFP-SMSr∆ SAM localized to the Golgi complex with residual amounts remaining in the ER. GFP-SMSr OD , on the other hand, displayed a subcellular distribution intermediate to that of GFP-SMSr and GFP-SMSr∆ SAM, with sizeable portions of the enzyme present in both ER and Golgi (Fig. 8a,b). This became even more apparent when we quantified the relative portion of GFP-tagged enzyme that co-localized with the Golgi marker GM130 using Manders' co-efficient 45 (Fig. 8c). Thus, mutations that perturb SMSr homo-oligomerization promote a redistribution of the enzyme from the ER to the Golgi complex. Interestingly, we found that treatment of cells with curcumin caused the opposite trend by promoting a redistribution of the enzyme from the Golgi to the ER (Fig. 8d). Together, these results indicate that SAM-mediated homo-oligomerization of SMSr serves a critical role in retaining the enzyme in the ER.

Discussion
Previous work revealed that SMSr, a CPE synthase with potential dual activity as ceramide sensor, requires its N-terminal SAM domain to control ER ceramide levels and suppress ceramide-induced mitochondrial apoptosis. In the present study, we uncovered a striking structural and functional similarity between SMSr-SAM and the SAM domain of the diacylglycerol kinase DGKδ . We show that SMSr-SAM, analogous to DGKδ -SAM, drives self-assembly of the enzyme into oligomers and that SAM-mediated oligomerization is a critical determinant of the enzyme's subcellular localization. Thus, our study provides the first example of a polytopic membrane protein that undergoes homotypic oligomerization via its SAM domain and indicates that self-association of SMSr is part of the mechanism by which the enzyme is retained in the ER.
Complementary lines of evidence indicate that the SAM domain of SMSr mediates self-assembly of the enzyme into oligomers. To begin with, SMSr-SAM is structurally and functionally related to the SAM domain of DGKδ , an enzyme that forms homotypic oligomers in the cytosol via its SAM domain to control its function as regulator of lipid signaling at the plasma membrane 8 . Residues critical for oligomerization of DGKδ -SAM are conserved in SMSr-SAM and substitution of these residues abolishes self-assembly of isolated SMSr-SAM in vitro and weakens the formation of SMSr homotypic oligomers in vivo. Moreover, chemical crosslinking studies indicate that SMSr self-assembles into trimers and hexamers, thus resembling the helical oligomers formed by DGKδ -SAM, which comprise six monomers per turn 6 . SMSr expressed in budding yeast, an organism that lacks structural homologs of SMS enzymes, forms higher-order complexes that co-migrate with SMSr trimers and hexamers assembled in human cells, in line with the idea that SMSr oligomerization involves self-assembly of the enzyme through its SAM domain and does not rely on any auxiliary protein.
Scientific RepoRts | 7:41290 | DOI: 10.1038/srep41290 SAM-mediated oligomerization of DGKδ prevents translocation of the enzyme to the plasma membrane, where its presence is required to attenuate DAG signaling and initiate PA signaling 6,8,46 . Substitution of key residues at the SAM oligomer interface abolishes DGKδ oligomerization and constitutively localizes the enzyme to the plasma membrane 6 . Analogous to the role of DGKδ -SAM, SAM-mediated oligomerization of SMSr influences its subcellular localization. While an oligomerization competent form of SMSr predominantly localizes to the ER, mutations that destabilize SMSr oligomers cause a partial redistribution of the enzyme to the Golgi. Conversely, treatment of cells with curcumin, a drug that stabilizes SMSr homo-oligomers, promotes retention of the enzyme in the ER. It should be noted that combining three point mutations that each disrupt self-assembly of SMSr-SAM domains in vitro proved insufficient to completely abolish the formation of SMSr trimers and hexamers in vivo. This likely explains why removal of the SAM domain results in a more pronounced redistribution of the enzyme to the Golgi than introduction of mutations that only partially undermine SMSr homo-oligomerization. Together, our data indicate that SAM-mediated oligomerization of SMSr serves a critical role in retaining the enzyme in the ER. This is a remarkable finding in view of the fact that oligomerization has been frequently identified as a prerequisite for efficient export of membrane proteins from the ER [47][48][49][50][51] . How oligomerization of SMSr prevents the enzyme from entering COPII vesicles that bud from the ER remains to be established. One possibility is that SMSr carries an ER-export signal that is masked by its homotypic oligomerization. Alternatively, SMSr oligomers, in contrast to monomers, interact with an ER-resident protein to actively retain them in the ER or simply fail to enter COPII vesicles because of their size or shape.  SMSr belongs to the SMS enzyme family, which includes the Golgi-resident SM synthase SMS1. The latter enzyme also contains an N-terminal SAM domain and possesses the ability to self-associate. However, both SMS1 self-assembly and its localization to the Golgi complex were completely independent of its SAM domain. Moreover, the SMS1-SAM domain lacks the basic assembly of interface residues that are critical for homo-oligomerization of SMSr-SAM and DGKδ -SAM. Thus, even though SMSr and SMS1 display a substantial degree of structural similarity, their SAM domains appear to serve distinct roles. We previously showed that swapping the entire N-terminal cytosolic tail of SMS1 for that of SMSr blocked export of the chimera from the ER 21 . In contrast, exchanging only the SAM domain did not prevent the enzyme from reaching the Golgi. We anticipate that the activity of SMSr-SAM as ER retention signal in the SMS1 chimera is overruled by an ER-export signal present downstream of its SAM domain. In support of this notion, adding a C-terminal KKXX ER-retrieval signal failed to retain SMS1 in the ER but readily restored the ER-residency of SMSrΔ SAM.
SMSr is a critical regulator of ER ceramides, at least in a variety of cultured mammalian cell lines 21 . The substantial rise in ER ceramides observed upon acute (siRNA-mediated) depletion of SMSr is difficult to account for by loss of the rather modest CPE-producing activity of the enzyme 21 . We recently reported that SMSr-catalyzed CPE production, although required, is not sufficient to suppress ceramide accumulation, ruling out metabolic conversion of ceramides as the primary mechanism by which SMSr controls ER ceramide levels 25 . Instead, we found that SMSr-mediated control over ER ceramides relies on its SAM domain; while removal of SAM did not affect SMSr-catalyzed CPE production, it abolished the ability of the enzyme to suppress ceramide accumulation even when SMSrΔ SAM is retained in the ER 25 . This implies that SMSr-SAM, besides trapping the enzyme in the ER, directly participates in the mechanism by which SMSr controls ER ceramide levels. Application of bifunctional lipid analogues in photo-affinity labeling experiments with recombinant SMSr-SAM did not reveal any particular affinity of this domain for any of the enzyme's lipid substrates or products. In addition, chemical crosslinking studies showed that catalytic activity of SMSr is dispensable for the ability of the enzyme to form homo-oligomers. Attempts to address whether fluctuations in ER ceramide levels influence the oligomeric state of SMSr are inconclusive. Our finding that curcumin stabilizes SMSr homo-oligomers most likely reflects a secondary effect of this compound as chemical crosslinker 44 rather than its ability to stimulate de novo ceramide production in the ER 33,34 . Whether ER ceramides have an impact on SMSr oligomerization remains to be established. Dynamic changes in the stoichiometry of membrane proteins are hard to resolve by bulk biochemical approaches like chemical crosslinking or co-immunoprecipitation analysis. Therefore, our ongoing efforts focus on the application of single-molecule fluorescence microscopy as complementary method to monitor SMSr oligomerization in intact cells 52 .

Methods
Fluorescence microscopy. HeLa cells were seeded on glass coverslips and transfected with HA-or GFP-tagged SMS constructs using Effectene. After 24 h, cells were fixed with 4% (w/v) paraformaldehyde in PBS, quenched in 25 mM NH 4 and permeabilized using PBS containing 0.1% (w/v) saponin and 0.2% (w/v) BSA. Coverslips were immunostained using primary and fluorophore-conjugated secondary antibodies essentially as in Vacaru et al. 21 . Coverslips were mounted using Prolong Gold Antifade Reagent (Molecular Probes). Images were captured using a confocal microscope (Olympus LSM FV1000) with two spectral and a single standard detector, a UPLSAPO 60x/NA 1.35 oil immersion objective (Olympus) and an Olympus laser box with AOTF laser combiner. GFP/FITC/Cy2 fluorophores were excited with a 488 nm Argon laser, Cy3 was excited with a 559 nm diode laser and Cy5 was excited with a 635 nm diode laser. Excitation light was reflected by a 405/488/559/635 nm dichroic mirror. Emitted light was collected using secondary dichroic mirrors SDM-560 and SDM-640 and a barrier filter BA 655-755 for GFP/FITC/Cy2, Cy3 and Cy5, respectively. To avoid crosstalk between imaging channels, images for different fluorophores were collected sequentially in a descending order of wavelength i.e. from long to short. Imaging of entire cells was done by acquisition of z series of 250 nm step size. Quantitation of Golgi-associated levels of GFP-tagged SMSr and calnexin was performed using ImageJ software (NIH). First, Golgi regions in transfected cells were determined using anti-GM130 immunostaining. Fluorescence intensity values from areas without any cells were taken as background values. After background subtraction, the ratio of GFP or calnexin signal intensity within the Golgi region (Manders' coefficient) were determined using the JaCoP plugin of ImageJ software.